A novel variable step size ICA adaptive algorithm
LI Xiao-ou · Journal of Circuits and Systems · 2004
Based on the expected square of estimation function, infeasible-interior-point method and steepest descent, a novel variable step size ICA adaptive algorithm is presented with the minimum mutual information. The misadjustment in the steady state happens during the computation of separating matrix by the fixed step size , the novel algorithm overcomes this problem, and has faster convergence at the early stage and lower residual error than the ICA based on the simulated annealing step size. In the meantime, it has the ability to track a changing mixing environment. Theoretical analysis and simulation results demonstrate that this algorithm may improve the adaptability of ICA effectively and perform blind signal separation accurately.